Tutorial - Simple Net: To begin, we will read a Bayes net stored on disk. Choose File → Open from the menu, and when the standard file-opening dialog box appears, use it to open the file called “Chest Clinic” in the Examples folder. A window will appear containing a Bayes net consisting of several connected nodes. This very simple net for diagnosing patients arriving at a clinic is a classic example often used to introduce Bayes nets.
You can prepare the net for inference by choosing Network → Compile from the menu, or by clicking the toolbar button (if they are gray, the net has already been compiled). When the compilation is complete, the default display style changes so that nodes are displayed with bar graphs showing the beliefs for each of their states. You can enter a finding (also known as an “observation”, or “evidence”) for a node by clicking on the name of the finding to the left of the bar graph. You can also enter a finding by right-clicking on the node and choosing Enter Finding.
When the finding is entered, the displays of all the nodes will be adjusted to account for it. For instance, putting an ‘abnormal’ finding for the ‘XRay Result’ node increases the belief that the patient has lung cancer from 5.5% to 48.9%, but then indicating that the patient has made a visit to Asia decreases that belief to 37.1%, because the abnormal XRay is partially explained away by a greater chance of Tuberculosis (which the patient could catch in Asia).
If you choose “Unknown” from the findings menu, then any finding for that node will be retracted, and if you choose “Likelihood” you will be queried for the probabilities of an uncertain finding (“virtual evidence”) for the node. You can also click directly on the name of the finding a second time to retract the finding.
Each time you enter or retract a finding, the beliefs of all the nodes will immediately be updated to account for the new information. If you wish updating to only occur when you do a Network → Update command, you can turn off the auto-update feature by toggling Network → Automatic Updating.
If you want to observe or change the conditional probabilities of a node (which express its relation with its parent nodes), select the node by clicking once on it, and then choose Table → View/Edit, or click on the toolbar button with the relation symbol: . A special window called the table dialog box will open which displays the probabilities in a table. The left-hand side of the table contains a vertical list of parent configurations, and for each configuration the right-hand side has a few probabilities, expressed as percentages.
Each column of the right-hand side corresponds to a different state of the node. So each number represents the conditional probability that the node takes on the state indicated by the column the number is in, given that the parents have the configuration indicated by the row the number is in. If the node is deterministic (e.g. the “TbOrCa” node), then the table dialog box will display a function table. This is the same as the conditional probability table, except that the right-hand side simply has the state of the node which is the function value for that parent configuration.
Tutorial - Complex Nets: To work with examples of more complex Bayes nets, open the file called “Alarm” (also in the medical domain), or the file called “HailFinder” (weather prediction). You can compile them and do inference in the same way as you did with Chest Clinic. With this example, you can click directly on the name of the finding a second time to retract the finding or hold down the shift key while clicking on the name of the finding to enter a negative finding.
If you have entered a number of findings for a particular case and wish to save them in their own file, choose Cases → Save Case As and enter the file name in the dialog box which appears. If you want to work on a new case choose Cases → Remove Findings, and then enter the new findings. To recover the original case, choose Cases → Get Case and pick its name from the dialog box which appears.
When you are done with a net window, you can get rid of
it by clicking the button in its title
bar, or by making it the active
window and then choosing File
→ Close.
Home > Getting Started > Quick Tour > Probabilistic Inference